next up previous
Next: Acknowledgements Up: eqns.html Previous: Total Least Squares

Parallel Implementation

The iterative algorithm takes about 25 min. on a Sun Sparc20for the images used in this study. On larger images, this algorithm may not be sufficiently fast. This prompted the implementation of this algorithm on a parallel platform, the Connection Machine CM5 (a 32-node parallel machine). The implementation used the multiblock PARTI primitives[] developed at the University of Maryland. The parallelization is rather simple since all the operations involved are on a pixel basis, and do not depend on any neighbourhoods. The speedup achieved by this implementation is a factor of three, approximately.



Generated by latex2html-95.1
Sat Jan 6 18:58:19 EST 1996